MS ABI: Use musttail for vtable thunks that pass arguments by value

This moves some memptr specific code into the generic thunk emission
codepath.

Fixes PR20053.

Reviewers: majnemer

Differential Revision: http://reviews.llvm.org/D4613

llvm-svn: 214004

3 files changed, 127 insertions, 18 deletions

diff --git a/clang/test/CodeGenCXX/microsoft-abi-byval-thunks.cpp b/clang/test/CodeGenCXX/microsoft-abi-byval-thunks.cpp
index 799521725c5..288f3c68ce1 100644
--- a/clang/test/CodeGenCXX/microsoft-abi-byval-thunks.cpp
+++ b/clang/test/CodeGenCXX/microsoft-abi-byval-thunks.cpp
@@ -1,4 +1,4 @@
-// RUN: not %clang_cc1 %s -fno-rtti -triple=i686-pc-win32 -emit-llvm -o /dev/null 2>&1 | FileCheck --check-prefix=CHECK32 %s
+// RUN: %clang_cc1 %s -fno-rtti -triple=i686-pc-win32 -emit-llvm -o - | FileCheck --check-prefix=CHECK32 %s
 // RUN: %clang_cc1 %s -fno-rtti -triple=x86_64-pc-win32 -emit-llvm -o - | FileCheck --check-prefix=CHECK64 %s
 
 namespace byval_thunk {
@@ -11,15 +11,103 @@ struct Agg {
 
 struct A { virtual void foo(Agg x); };
 struct B { virtual void foo(Agg x); };
-struct C : A, B { virtual void foo(Agg x); };
-C c;
+struct C : A, B { C(); virtual void foo(Agg x); };
+C::C() {} // force emission
 
-// CHECK32: cannot compile this non-trivial argument copy for thunk yet
+// CHECK32-LABEL: define linkonce_odr x86_thiscallcc void @"\01?foo@C@byval_thunk@@W3AEXUAgg@2@@Z"
+// CHECK32:             (%"struct.byval_thunk::C"* %this, <{ %"struct.byval_thunk::Agg" }>* inalloca)
+// CHECK32:   %2 = getelementptr i8* %{{.*}}, i32 -4
+// CHECK32:   musttail call x86_thiscallcc void @"\01?foo@C@byval_thunk@@UAEXUAgg@2@@Z"
+// CHECK32:       (%"struct.byval_thunk::C"* %{{.*}}, <{ %"struct.byval_thunk::Agg" }>* inalloca %0)
+// CHECK32-NEXT: ret void
 
 // CHECK64-LABEL: define linkonce_odr void @"\01?foo@C@byval_thunk@@W7EAAXUAgg@2@@Z"
 // CHECK64:             (%"struct.byval_thunk::C"* %this, %"struct.byval_thunk::Agg"* %x)
 // CHECK64:   getelementptr i8* %{{.*}}, i32 -8
-// CHECK64:   call void @"\01?foo@C@byval_thunk@@UEAAXUAgg@2@@Z"(%"struct.byval_thunk::C"* %{{.*}}, %"struct.byval_thunk::Agg"* %x)
+// CHECK64:   call void @"\01?foo@C@byval_thunk@@UEAAXUAgg@2@@Z"
+// CHECK64:       (%"struct.byval_thunk::C"* %{{.*}}, %"struct.byval_thunk::Agg"* %x)
 // CHECK64-NOT: call
 // CHECK64:   ret void
 }
+
+namespace stdcall_thunk {
+struct Agg {
+  Agg();
+  Agg(const Agg &);
+  ~Agg();
+  int x;
+};
+
+struct A { virtual void __stdcall foo(Agg x); };
+struct B { virtual void __stdcall foo(Agg x); };
+struct C : A, B { C(); virtual void __stdcall foo(Agg x); };
+C::C() {} // force emission
+
+// CHECK32-LABEL: define linkonce_odr x86_stdcallcc void @"\01?foo@C@stdcall_thunk@@W3AGXUAgg@2@@Z"
+// CHECK32:             (<{ %"struct.stdcall_thunk::C"*, %"struct.stdcall_thunk::Agg" }>* inalloca)
+// CHECK32:   %[[this_slot:[^ ]*]] = getelementptr inbounds <{ %"struct.stdcall_thunk::C"*, %"struct.stdcall_thunk::Agg" }>* %0, i32 0, i32 0
+// CHECK32:   load %"struct.stdcall_thunk::C"** %[[this_slot]]
+// CHECK32:   getelementptr i8* %{{.*}}, i32 -4
+// CHECK32:   store %"struct.stdcall_thunk::C"* %{{.*}}, %"struct.stdcall_thunk::C"** %[[this_slot]]
+// CHECK32:   musttail call x86_stdcallcc void @"\01?foo@C@stdcall_thunk@@UAGXUAgg@2@@Z"
+// CHECK32:       (<{ %"struct.stdcall_thunk::C"*, %"struct.stdcall_thunk::Agg" }>*  inalloca %0)
+// CHECK32-NEXT: ret void
+
+// CHECK64-LABEL: define linkonce_odr void @"\01?foo@C@stdcall_thunk@@W7EAAXUAgg@2@@Z"
+// CHECK64:             (%"struct.stdcall_thunk::C"* %this, %"struct.stdcall_thunk::Agg"* %x)
+// CHECK64:   getelementptr i8* %{{.*}}, i32 -8
+// CHECK64:   call void @"\01?foo@C@stdcall_thunk@@UEAAXUAgg@2@@Z"
+// CHECK64:       (%"struct.stdcall_thunk::C"* %{{.*}}, %"struct.stdcall_thunk::Agg"* %x)
+// CHECK64-NOT: call
+// CHECK64:   ret void
+}
+
+namespace sret_thunk {
+struct Agg {
+  Agg();
+  Agg(const Agg &);
+  ~Agg();
+  int x;
+};
+
+struct A { virtual Agg __cdecl foo(Agg x); };
+struct B { virtual Agg __cdecl foo(Agg x); };
+struct C : A, B { C(); virtual Agg __cdecl foo(Agg x); };
+C::C() {} // force emission
+
+// CHECK32-LABEL: define linkonce_odr %"struct.sret_thunk::Agg"* @"\01?foo@C@sret_thunk@@W3AA?AUAgg@2@U32@@Z"
+// CHECK32:             (<{ %"struct.sret_thunk::C"*, %"struct.sret_thunk::Agg"*, %"struct.sret_thunk::Agg" }>* inalloca)
+// CHECK32:   %[[this_slot:[^ ]*]] = getelementptr inbounds <{ %"struct.sret_thunk::C"*, %"struct.sret_thunk::Agg"*, %"struct.sret_thunk::Agg" }>* %0, i32 0, i32 0
+// CHECK32:   load %"struct.sret_thunk::C"** %[[this_slot]]
+// CHECK32:   getelementptr i8* %{{.*}}, i32 -4
+// CHECK32:   store %"struct.sret_thunk::C"* %{{.*}}, %"struct.sret_thunk::C"** %[[this_slot]]
+// CHECK32:   %[[rv:[^ ]*]] = musttail call %"struct.sret_thunk::Agg"* @"\01?foo@C@sret_thunk@@UAA?AUAgg@2@U32@@Z"
+// CHECK32:       (<{ %"struct.sret_thunk::C"*, %"struct.sret_thunk::Agg"*, %"struct.sret_thunk::Agg" }>*  inalloca %0)
+// CHECK32-NEXT: ret %"struct.sret_thunk::Agg"* %[[rv]]
+
+// CHECK64-LABEL: define linkonce_odr void @"\01?foo@C@sret_thunk@@W7EAA?AUAgg@2@U32@@Z"
+// CHECK64:             (%"struct.sret_thunk::C"* %this, %"struct.sret_thunk::Agg"* noalias sret %agg.result, %"struct.sret_thunk::Agg"* %x)
+// CHECK64:   getelementptr i8* %{{.*}}, i32 -8
+// CHECK64:   call void @"\01?foo@C@sret_thunk@@UEAA?AUAgg@2@U32@@Z"
+// CHECK64:       (%"struct.sret_thunk::C"* %{{.*}}, %"struct.sret_thunk::Agg"* sret %agg.result, %"struct.sret_thunk::Agg"* %x)
+// CHECK64-NOT: call
+// CHECK64:   ret void
+}
+
+#if 0
+// FIXME: When we extend LLVM IR to allow forwarding of varargs through musttail
+// calls, use this test.
+namespace variadic_thunk {
+struct Agg {
+  Agg();
+  Agg(const Agg &);
+  ~Agg();
+  int x;
+};
+
+struct A { virtual void foo(Agg x, ...); };
+struct B { virtual void foo(Agg x, ...); };
+struct C : A, B { C(); virtual void foo(Agg x, ...); };
+C::C() {} // force emission
+}
+#endif
diff --git a/clang/test/CodeGenCXX/microsoft-abi-nontrivial-covariant-thunk.cpp b/clang/test/CodeGenCXX/microsoft-abi-nontrivial-covariant-thunk.cpp
index d305dd83271..cb8b5224d65 100644
--- a/clang/test/CodeGenCXX/microsoft-abi-nontrivial-covariant-thunk.cpp
+++ b/clang/test/CodeGenCXX/microsoft-abi-nontrivial-covariant-thunk.cpp
@@ -18,7 +18,7 @@ struct B {
 struct C : A, B {
   C();
   int c;
-  virtual C *clone(A); // expected-error {{cannot compile this non-trivial argument copy for thunk yet}}
+  virtual C *clone(A); // expected-error {{cannot compile this non-trivial argument copy for return-adjusting thunk yet}}
 };
 B::B() {}  // force emission
 C::C() {}  // force emission
diff --git a/clang/test/CodeGenCXX/microsoft-abi-virtual-member-pointers.cpp b/clang/test/CodeGenCXX/microsoft-abi-virtual-member-pointers.cpp
index 974953c0144..ffad7e70ce5 100644
--- a/clang/test/CodeGenCXX/microsoft-abi-virtual-member-pointers.cpp
+++ b/clang/test/CodeGenCXX/microsoft-abi-virtual-member-pointers.cpp
@@ -18,6 +18,7 @@ struct C {
   virtual int bar(int, double);
   virtual S baz(int);
   virtual S qux(U);
+  virtual S __fastcall zed(U);
 };
 
 namespace {
@@ -43,6 +44,9 @@ void f() {
   S (C::*ptr5)(U);
   ptr5 = &C::qux;
 
+  S (__fastcall C::*ptr6)(U);
+  ptr6 = &C::zed;
+
 
 // CHECK32-LABEL: define void @"\01?f@@YAXXZ"()
 // CHECK32: store i8* bitcast (void (%struct.C*)* @"\01??_9C@@$BA@AE" to i8*), i8** %ptr
@@ -64,14 +68,14 @@ void f() {
 // CHECK32-LABEL: define linkonce_odr x86_thiscallcc void @"\01??_9C@@$BA@AE"(%struct.C* %this) unnamed_addr
 // CHECK32: [[VPTR:%.*]] = getelementptr inbounds void (%struct.C*)** %{{.*}}, i64 0
 // CHECK32: [[CALLEE:%.*]] = load void (%struct.C*)** [[VPTR]]
-// CHECK32: musttail call x86_thiscallcc void [[CALLEE]](%struct.C* %{{.*}})
+// CHECK32: call x86_thiscallcc void [[CALLEE]](%struct.C* %{{.*}})
 // CHECK32: ret void
 // CHECK32: }
 //
 // CHECK64-LABEL: define linkonce_odr void @"\01??_9C@@$BA@AA"(%struct.C* %this) unnamed_addr
 // CHECK64: [[VPTR:%.*]] = getelementptr inbounds void (%struct.C*)** %{{.*}}, i64 0
 // CHECK64: [[CALLEE:%.*]] = load void (%struct.C*)** [[VPTR]]
-// CHECK64: musttail call void [[CALLEE]](%struct.C* %{{.*}})
+// CHECK64: call void [[CALLEE]](%struct.C* %{{.*}})
 // CHECK64: ret void
 // CHECK64: }
 
@@ -79,14 +83,14 @@ void f() {
 // CHECK32-LABEL: define linkonce_odr x86_thiscallcc i32 @"\01??_9C@@$B3AE"(%struct.C* %this, i32, double) unnamed_addr
 // CHECK32: [[VPTR:%.*]] = getelementptr inbounds i32 (%struct.C*, i32, double)** %{{.*}}, i64 1
 // CHECK32: [[CALLEE:%.*]] = load i32 (%struct.C*, i32, double)** [[VPTR]]
-// CHECK32: [[CALL:%.*]] = musttail call x86_thiscallcc i32 [[CALLEE]](%struct.C* %{{.*}}, i32 %{{.*}}, double %{{.*}})
+// CHECK32: [[CALL:%.*]] = call x86_thiscallcc i32 [[CALLEE]](%struct.C* %{{.*}}, i32 %{{.*}}, double %{{.*}})
 // CHECK32: ret i32 [[CALL]]
 // CHECK32: }
 //
 // CHECK64-LABEL: define linkonce_odr i32 @"\01??_9C@@$B7AA"(%struct.C* %this, i32, double) unnamed_addr
 // CHECK64: [[VPTR:%.*]] = getelementptr inbounds i32 (%struct.C*, i32, double)** %{{.*}}, i64 1
 // CHECK64: [[CALLEE:%.*]] = load i32 (%struct.C*, i32, double)** [[VPTR]]
-// CHECK64: [[CALL:%.*]] = musttail call i32 [[CALLEE]](%struct.C* %{{.*}}, i32 %{{.*}}, double %{{.*}})
+// CHECK64: [[CALL:%.*]] = call i32 [[CALLEE]](%struct.C* %{{.*}}, i32 %{{.*}}, double %{{.*}})
 // CHECK64: ret i32 [[CALL]]
 // CHECK64: }
 
@@ -94,14 +98,14 @@ void f() {
 // CHECK32-LABEL: define linkonce_odr x86_thiscallcc void @"\01??_9C@@$B7AE"(%struct.C* %this, %struct.S* noalias sret %agg.result, i32) unnamed_addr
 // CHECK32: [[VPTR:%.*]] = getelementptr inbounds void (%struct.C*, %struct.S*, i32)** %{{.*}}, i64 2
 // CHECK32: [[CALLEE:%.*]] = load void (%struct.C*, %struct.S*, i32)** [[VPTR]]
-// CHECK32: musttail call x86_thiscallcc void [[CALLEE]](%struct.C* %{{.*}}, %struct.S* sret %agg.result, i32 %{{.*}})
+// CHECK32: call x86_thiscallcc void [[CALLEE]](%struct.C* %{{.*}}, %struct.S* sret %agg.result, i32 %{{.*}})
 // CHECK32: ret void
 // CHECK32: }
 //
 // CHECK64-LABEL: define linkonce_odr void @"\01??_9C@@$BBA@AA"(%struct.C* %this, %struct.S* noalias sret %agg.result, i32) unnamed_addr
 // CHECK64: [[VPTR:%.*]] = getelementptr inbounds void (%struct.C*, %struct.S*, i32)** %{{.*}}, i64 2
 // CHECK64: [[CALLEE:%.*]] = load void (%struct.C*, %struct.S*, i32)** [[VPTR]]
-// CHECK64: musttail call void [[CALLEE]](%struct.C* %{{.*}}, %struct.S* sret %agg.result, i32 %{{.*}})
+// CHECK64: call void [[CALLEE]](%struct.C* %{{.*}}, %struct.S* sret %agg.result, i32 %{{.*}})
 // CHECK64: ret void
 // CHECK64: }
 
@@ -109,28 +113,45 @@ void f() {
 // CHECK32-LABEL: define internal x86_thiscallcc void @"\01??_9D@?A@@$BA@AE"(%"struct.(anonymous namespace)::D"* %this) unnamed_addr
 // CHECK32: [[VPTR:%.*]] = getelementptr inbounds void (%"struct.(anonymous namespace)::D"*)** %{{.*}}, i64 0
 // CHECK32: [[CALLEE:%.*]] = load void (%"struct.(anonymous namespace)::D"*)** [[VPTR]]
-// CHECK32: musttail call x86_thiscallcc void [[CALLEE]](%"struct.(anonymous namespace)::D"* %{{.*}})
+// CHECK32: call x86_thiscallcc void [[CALLEE]](%"struct.(anonymous namespace)::D"* %{{.*}})
 // CHECK32: ret void
 // CHECK32: }
 //
 // CHECK64-LABEL: define internal void @"\01??_9D@?A@@$BA@AA"(%"struct.(anonymous namespace)::D"* %this) unnamed_addr
 // CHECK64: [[VPTR:%.*]] = getelementptr inbounds void (%"struct.(anonymous namespace)::D"*)** %{{.*}}, i64 0
 // CHECK64: [[CALLEE:%.*]] = load void (%"struct.(anonymous namespace)::D"*)** [[VPTR]]
-// CHECK64: musttail call void [[CALLEE]](%"struct.(anonymous namespace)::D"* %{{.*}})
+// CHECK64: call void [[CALLEE]](%"struct.(anonymous namespace)::D"* %{{.*}})
 // CHECK64: ret void
 // CHECK64: }
 
-// Thunk for calling the fourth virtual function in C, taking a struct parameter and returning a struct.
+// Thunk for calling the fourth virtual function in C, taking a struct parameter
+// and returning a struct.
 // CHECK32-LABEL: define linkonce_odr x86_thiscallcc %struct.S* @"\01??_9C@@$BM@AE"(%struct.C* %this, <{ %struct.S*, %struct.U }>* inalloca) unnamed_addr
 // CHECK32: [[VPTR:%.*]] = getelementptr inbounds %struct.S* (%struct.C*, <{ %struct.S*, %struct.U }>*)** %{{.*}}, i64 3
 // CHECK32: [[CALLEE:%.*]] = load %struct.S* (%struct.C*, <{ %struct.S*, %struct.U }>*)** [[VPTR]]
-// CHECK32: [[CALL:%.*]] = musttail call x86_thiscallcc %struct.S* [[CALLEE]](%struct.C* %this, <{ %struct.S*, %struct.U }>* inalloca %{{.*}})
-// CHECK32: ret %struct.S* [[CALL]]
+// CHECK32: [[CALL:%.*]] = musttail call x86_thiscallcc %struct.S* [[CALLEE]](%struct.C* %{{.*}}, <{ %struct.S*, %struct.U }>* inalloca %{{.*}})
+// CHECK32-NEXT: ret %struct.S* [[CALL]]
 // CHECK32: }
 //
 // CHECK64-LABEL: define linkonce_odr void @"\01??_9C@@$BBI@AA"(%struct.C* %this, %struct.S* noalias sret %agg.result, %struct.U*) unnamed_addr
 // CHECK64: [[VPTR:%.*]] = getelementptr inbounds void (%struct.C*, %struct.S*, %struct.U*)** %{{.*}}, i64 3
 // CHECK64: [[CALLEE:%.*]] = load void (%struct.C*, %struct.S*, %struct.U*)** [[VPTR]]
-// CHECK64: musttail call void [[CALLEE]](%struct.C* %this, %struct.S* sret %agg.result, %struct.U* %{{.*}})
+// CHECK64: call void [[CALLEE]](%struct.C* %{{.*}}, %struct.S* sret %agg.result, %struct.U* %{{.*}})
+// CHECK64: ret void
+// CHECK64: }
+
+// Thunk for calling the fifth virtual function in C, taking a struct parameter
+// and returning a struct.
+// CHECK32-LABEL: define linkonce_odr x86_fastcallcc void @"\01??_9C@@$BBA@AE"(%struct.C* inreg %this, %struct.S* inreg noalias sret %agg.result, <{ %struct.U }>* inalloca) unnamed_addr
+// CHECK32: [[VPTR:%.*]] = getelementptr inbounds void (%struct.C*, %struct.S*, <{ %struct.U }>*)** %{{.*}}, i64 4
+// CHECK32: [[CALLEE:%.*]] = load void (%struct.C*, %struct.S*, <{ %struct.U }>*)** [[VPTR]]
+// CHECK32: musttail call x86_fastcallcc void [[CALLEE]](%struct.C* inreg %{{.*}}, %struct.S* inreg sret %{{.*}}, <{ %struct.U }>* inalloca %{{.*}})
+// CHECK32-NEXT: ret void
+// CHECK32: }
+//
+// CHECK64-LABEL: define linkonce_odr void @"\01??_9C@@$BCA@AA"(%struct.C* %this, %struct.S* noalias sret %agg.result, %struct.U*) unnamed_addr
+// CHECK64: [[VPTR:%.*]] = getelementptr inbounds void (%struct.C*, %struct.S*, %struct.U*)** %{{.*}}, i64 4
+// CHECK64: [[CALLEE:%.*]] = load void (%struct.C*, %struct.S*, %struct.U*)** [[VPTR]]
+// CHECK64: call void [[CALLEE]](%struct.C* %{{.*}}, %struct.S* sret %agg.result, %struct.U* %{{.*}})
 // CHECK64: ret void
 // CHECK64: }